Research map / Learning theory and algorithms / Sample complexity
Estimation of the Order of Non-Parametric Hidden Markov Models using the Singular Values of an Integral Operator
Marie Du Roy de Chaumaray, Salima El Kolei, Marie-Pierre Etienne, Matthieu Marbac
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use proposed integral-operator estimator instead of spectral method with slope heuristic | HMM order estimation · Gaussian emissions, n=250 | % selecting true order: 60 vs 4 (percent selecting L=3) | Improved |
| use proposed estimator instead of spectral method with kmeans-tuned threshold | HMM order estimation · Gaussian emissions, n=500 | % selecting true order: 77 vs 47 (percent selecting L=3) | Improved |
| tune spectral threshold constant with mini-batch Kmeans procedure instead of slope heuristic | spectral method · Gaussian emissions, n=4000 | % selecting true order: 15 -> 100 (percent selecting L=3) | Improved |
| increase sample size from 250 to 1000 | proposed estimator · Gaussian emissions, 5% error rate | % selecting true order: 60 -> 100 | Improved |
| increase sample size from 250 to 500 | proposed estimator · Student emissions, 5% error rate | % selecting true order: 63 -> 85 | Improved |
| use heavier-tailed Gaussian vs Laplace emissions at small n | proposed estimator · n=250, 5% error rate | % selecting true order: Gaussian 60, Student 63, Laplace 81, Von-Mises 89 | Mixed |
| small sample size causes underestimation, not overestimation | proposed estimator · Gaussian emissions, n=250 | % selecting L<3: 40% selected L=2 and 0% L>3 | Hurt |
| use proposed method instead of AIC on booby trajectory (step length, turning angle) | order selection · Booby Trip 1 | estimated order: 10 -> 3 | Improved |
| use proposed method instead of BIC on booby trajectory (velocity, rotational velocity) | order selection · Booby Trip 1 | estimated order: 9 -> 3 | Improved |
| use proposed method instead of ICL on booby trajectory (step length, turning angle) | order selection · Booby Trip 2 | estimated order: 6 -> 3 | Improved |
| use velocity and rotational velocity instead of step length and turning angle metrics | proposed method · Booby trajectories, adjusted Rand index between latent sequences | adjusted Rand index: 0.959, 0.751 and 0.980 for trips 1, 2 and 3 | No clear effect |
Closest papers by meaning
- Learning HMMs with Nonparametric Emissions via Spectral Decompositions of Continuous Matrices (NeurIPS 2016 · similarity 0.88)
- Minimax Adaptive Estimation of Nonparametric Hidden Markov Models (JMLR 2016 · similarity 0.88)
- Determine the Number of States in Hidden Markov Models via Marginal Likelihood (JMLR 2025 · similarity 0.87)
- Identifiability and Consistent Estimation of Nonparametric Translation Hidden Markov Models with General State Space (JMLR 2020 · similarity 0.86)
- A Spectral Algorithm for Inference in Hidden semi-Markov Models (JMLR 2017 · similarity 0.86)
- State-by-state Minimax Adaptive Estimation for Nonparametric Hidden {M}arkov Models (JMLR 2018 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
